In the three weeks since GPT-5.5 launched, it’s been a tale of two cities. While the finance crowd has fallen head over heels for Anthropic’s growth and impending October IPO, AI engineers have shown a marked swing toward pro-Codex sentiment. The shift appears driven by three factors: GPT-103 proving to be an exceptionally strong model (in some cases reaching Mythos-tier performance), the launch of Codex for Everything Else, and—most crucially for today’s piece—significantly more generous usage limits.
Claude’s pricing announcement itself was handled fairly well, but it simply wasn’t what users of alternative harnesses wanted to hear: every Claude subscription now includes a monthly API token credit equal to the dollar value of the plan. So for $200, you get a Claude subscription that comes with its own separate usage limits when accessing Claude on Anthropic’s own platforms. AI and Claude Code for interactive usage, plus $200 in API credits for all other Claude applications like Claude-P, OpenClaw, and more for programmatic usage. If this structure had been in place from the beginning, it would have been considered an excellent deal. However, given the historical subsidy and pricing advantages (estimated at a 70-90% discount to regular API pricing), many are seeing it as a kind of „rug pull“—though it’s good to finally have a clear official policy instead of the previous selective enforcement against OpenClaw, OpenCode, and the uncertain status of less popular harnesses. That these announcements dropped on the same day OpenAI launched its enterprise-switch promotion is an incredible coincidence. In the end, we’d advise against reading too much into short-term price fluctuations in either direction. Both labs are performing strongly, and these adjustments are, in the broader context, typical pricing experiments as they pioneer the future of coding and determine the right pricing model while disrupting a decades-old industry. **Paraphrased:**
Anthropic started out more permissive, but now that Claude Code has established a strong brand and reputation as an agent platform, they’re prioritizing their best pricing for their own tools and metering everything else. In contrast, Codex—as the challenger—is being far more generous across the board. Maybe hardware is destiny. Or maybe this is just part of the recurring six-month “mandate equinox” cycle.
dan@irl_danB. Wow, almost six months ahead of schedule (within three weeks), the mandate equinox is real—Anthropic will recapture hearts and minds around October, right on time for recursive self-improvement. dan @irl_danB.
Latent.Space